Comparative Analysis of Spatial Image Classification Techniques Using Rough-Sets

Abstract

Many traditional spatial image classification techniques which are developed over past years and exists in literature like maximum likelihood, minimum distance, and parallelepiped suffer from problem of misclassification. Today, expert systems along with machine learning methods are getting universality in this area due to effective classification. This paper emphasizes on Artificial Neural Network (ANN) and “Rough Set” based “Artificial Neural Network” (RS-ANN) methods. In RS-ANN technique, Rough set (RS) is used as an attributes selection mathematical tool which excludes the redundant attributes. Further, this reduced dimensionality data set is given to Artificial Neural Network (ANN) classifier correspondingly. This process improves the classification accuracy and performance. The experiments are performed with LISS III spatial images of Hyderabad region collected from NRSC. The comparative analysis is carried out between ANN and RS-ANN techniques by considering efficiency measures like root mean square error and execution time. It is observed that RS-ANN technique outperforms the ANN technique for the spatial images with respect to accuracy and performance.